ResearchTasksPrefill-stage hallucination risk detectionPrefill-stage hallucination risk detectionFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedBenchmark-500 Relaxed Consensus (Pvote ≥ 0.8)Risk-Cos0.696AUROC (Mean)4Mar 23, 2026Benchmark-500 Strict Consensus Pvote = 1.0 vs. CleanRisk-Cos0.694AUROC (Mean)4Mar 23, 2026